Multiprocessing error when running mute slam on AGX orin

[yhaddouda]

Hello,
I am trying to run your code on AGX Orin 64GB, all the dependencies seem to be installed successfully, when I run the code I get this error :
Traceback (most recent call last):
File "", line 1, in
File "/home/yh279050/miniforge3/envs/mute_slam/lib/python3.10/multiprocessing/spawn.py", line 116, in spawn_main
exitcode = _main(fd, parent_sentinel)
File "/home/yh279050/miniforge3/envs/mute_slam/lib/python3.10/multiprocessing/spawn.py", line 126, in _main
self = reduction.pickle.load(from_parent)
File "/home/yh279050/miniforge3/envs/mute_slam/lib/python3.10/site-packages/torch/multiprocessing/reductions.py", line 148, in rebuild_cuda_tensor
storage = storage_cls._new_shared_cuda(
File "/home/yh279050/miniforge3/envs/mute_slam/lib/python3.10/site-packages/torch/storage.py", line 1157, in _new_shared_cuda
return torch.UntypedStorage._new_shared_cuda(*args, **kwargs)
RuntimeError: CUDA error: invalid argument
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Have you tried running the code on this platform by any chance? and if not is there a version of your code that doesn't use multiprocessing ?

[lumennYan]

Hello, I haven't tried running the code on AGX Orin 64GB. Since the code is modified from NICE-SLAM, which uses multiprocessing, I haven't implentment a different version. But I think you can easily disable multiprocessing by putting Tracking and Mapping in the same process, since they run in a mutually blocking way and only communicate with each other through image index and submap list.